Record — capture both sides
Microphone, device audio, or both at once as two separate tracks. 5-second chunks hit the disk as you talk, so a crash can't eat your meeting.
Every meeting.Every word.
Never leaves your machine.
Local Transcriber records your calls, transcribes them with whisper.cpp or Parakeet on your own GPU, and turns them into summaries your team can use. No account. No cloud. No uploads.
Fifteen seconds. The whole workflow.
Record → transcribe → share. Press play.
01RecordMic + call, two tracks
02TranscribeOn your GPU
03ShareIssues, calendar, export
Record first. Transcribe when you're ready.
Recording and transcription are separate on purpose. Your audio is saved the moment it's captured — AI is an optional second step that can never damage the original.
Transcribe — native, GPU-fast
One click runs whisper.cpp or Parakeet locally, with live progress, cancel and retry. Language is auto-detected. Nothing is uploaded.
Share — summarize & ship
A local LLM writes the summary and key points. Push it to a GitLab wiki, draft follow-up events for your calendar, or export TXT, Markdown and JSON.
Everything a meeting needs.
Nothing it doesn't. A desktop app built on Tauri and Rust — small, fast and native on every platform.
Two-way capture for online calls
Your mic and the call audio are recorded as two independent tracks, so every line is labeled Me or Others — speaker labels without diarization guesswork.
One-click GPU
Vulkan and CUDA acceleration through voxtype. Metal on Apple Silicon.
Crash-safe by design
Audio streams to disk in small chunks. Power cut? Reopen the app and hit Recover.
Tiered model catalog
Models ranked S–D by accuracy. Pick, download, done.
Full-text search
Find that one sentence across every meeting.
Summaries from a local LLM
Use Ollama or an OpenCode agent on your machine, or point it at any OpenAI-compatible endpoint you choose.
Your company calendar
CalDAV — SOGo, Nextcloud, Zimbra, Kerio — plus Microsoft Graph and Exchange EWS. Follow-ups stay drafts until you click Approve.
Team sharing without a backend
Publish transcripts and summaries to a GitLab wiki, issue or repository folder. Access follows project membership.
Install once, stay up to date
One click on Update & restart downloads, verifies and installs — never without your click, never mid-recording.
Make it yours.
Pick a brand colour and an accent. The app has the same two controls in Settings → Appearance, and the whole interface follows. Try it here — this page changes with you.
Brand#f48ca8
Accent#ff4e7e
[x] 01 Transcribe02 GitLab03 Calendar04 Export
[x] Finish crash-recovery QA qa
[x] Ship GitLab export · one folder per meeting feature
[ ] Release candidate for offline mode release
Private by architecture, not by promise.
There's no account and no server to trust. Here is exactly what can leave your machine — and when.
| Data | What happens | Leaves your machine? |
|---|---|---|
| Your audio | Recorded, stored and transcribed locally. | [x] Never |
| Transcripts & summaries | Local by default — with Ollama, even the AI runs on-device. | [x] Stays local |
| Analytics & telemetry | None. Not anonymous, not opt-out. None. | [x] None |
| Update check | On launch, asks GitHub if a newer version exists. Sends nothing about you. Off with one switch. | Your choice |
| GitLab, calendar, remote LLM | Only if you turn them on, only to endpoints you configure. | Your choice |
0accounts
0servers of ours
0bytes of audio uploaded
The full details are in the privacy policy.
Download
Click download, open it, record. Transcription is built in — free, MIT-licensed, no sign-up.
LatestChecking GitHub…
On a phone? Local Transcriber runs on your computer. Pick its system below, or send yourself the link.
No account, no setup wizard. New versions install from inside the app in one click.
whisper.cpp ships inside the app. The speech model downloads once, on your first transcription.
Every binary lists its SHA-256. Compare with
shasum -a 256 <file>.—
Downloads and in-app updates of all releases, as counted by GitHub Releases — the public total across every version and platform. No cookies, no IDs, no personal data. Just a number.
# build it yourself $ git clone https://github.com/GrzeskoByte/local-transcript $ cd local-transcript $ npm install --ignore-scripts $ npm run tauri:build # installers → src-tauri/target/release/bundle/
# optional: bring your own GPU engine # Linux — voxtype with Vulkan/CUDA $ sudo voxtype setup gpu --enable # Windows — CUDA/Vulkan whisper.cpp build > setx WHISPER_CLI_PATH C:\whisper\whisper-cli.exe
FAQ
Does my audio ever get uploaded?
No. Recording, storage and transcription all happen on your computer. The only features that send anything over the network are ones you switch on yourself — GitLab sharing, calendar sync or a remote LLM — and they send text, never audio.
Do I need to install anything else?
No. The transcription engine (whisper.cpp) ships inside the app. The first time you click Transcribe, it downloads its speech model once (about 550 MB) with a progress bar — after that everything works offline. Power users can plug in their own GPU engine (voxtype or a whisper.cpp build) and the app will use it automatically.
Do I need an internet connection?
Only to download the app and the speech model once. After that, recording, transcription and local summaries work fully offline. The calendar keeps each month on your device too, so it opens instantly and still shows your events offline.
How do I update?
You don't have to come back here. When a new version is out, a notice appears in the sidebar — click Update & restart. The download is signature-checked before it installs, and updating is paused while you're recording or transcribing. You can also check by hand or turn off the launch check in Settings → App → Updates. On Linux the AppImage updates itself; .deb and .rpm installs get a Download button instead. Versions before 0.1.2 need one manual install to get the updater.
Which calendars work?
Anything that speaks CalDAV — SOGo, Nextcloud, Zimbra, Kerio and others — plus Microsoft 365 (Graph) and on-prem Exchange (EWS). Your mail settings (IMAP/SMTP) don't carry calendars, so Settings offers shortcuts: Detect from server tries the usual addresses and lists your calendars, Find in Thunderbird reuses the calendars Thunderbird already knows (without passwords), and Use SOGo address fills it in from your host and username. A meeting can have several follow-up events, and each one is a draft until you approve it.
Which speech model should I use?
Start with large-v3-turbo — it's the default and a great balance of speed and accuracy. For the best accuracy, parakeet-tdt-0.6b-v3 sits at the top of the S-tier; on Linux it needs voxtype's ONNX variant. Settings ranks every model so the choice is obvious.
Can it record Zoom, Meet or Teams calls?
Yes — choose Mic + Device and share the call's tab or window when asked. Your voice and the call are saved as separate tracks. Device-audio availability depends on your OS, and headphones are recommended so the mic doesn't pick up the speakers.
What happens if transcription fails?
Nothing bad. The recording is the source of truth and is never modified by transcription. You'll see "Your original recording is still safe." and a Retry button.
macOS says the app can't be opened — what now?
Early builds may not be notarized yet. Right-click the app in Applications and choose Open, or run xattr -dr com.apple.quarantine "/Applications/Local Transcribe.app".
Is it really free?
Yes. Local Transcriber is open source under the MIT license. No tiers, no trial, no account.
More questions? The documentation has a full user manual — installation, recording, transcription, integrations and troubleshooting, with screenshots.